A new detection method models text as a graph of sentence transitions, exposing the tell-tale over-regularity of AI-generated writing. The approach flags synthetic prose by measuring how unnaturally smooth its flow of ideas is compared to human text.
A new benchmark, Multi2AV-Safety, evaluates safety risks in multimodal-to-audio-video generation systems—probing how text, image, and audio prompts can produce harmful synthetic media across the AV generation stack.
A technical breakdown of Flow Matching, Optimal Transport, and Schrödinger bridges — the mathematical engines increasingly powering the generative models behind modern AI image and video synthesis.
Google DeepMind has launched the world's first double-blind AI evaluations, a new methodology designed to remove bias from model testing and improve trust in AI safety and capability assessments.
Behind every conversational voice AI product lies a hidden architecture battle over latency. The difference between 2000ms and 250ms determines whether an AI voice feels robotic or human — and it's driving a fundamental redesign of synthetic speech pipelines.
Zero-knowledge proofs are emerging as a cryptographic backbone for verifying AI outputs without exposing models or data—a foundational tool for digital authenticity, model provenance, and trust in an era of synthetic media.
Major AI models are increasingly embedding invisible statistical watermarks into the text they generate. Here's how these hidden signals work, why they matter for digital authenticity, and what they mean for detecting synthetic content.
LLM watermarking promises to help detect AI-generated text, but embedding these statistical signatures comes at a cost. New analysis shows how watermarking degrades output quality, reasoning, and reliability, raising hard questions for digital authenticity efforts.
A 21-year-old founder is developing a deepfake audio detection startup that alerts users in real time during live phone and video calls, targeting the fast-rising threat of voice-cloning scams.
Alibaba has rolled out Wan3.0, the newest iteration of its AI video generation model, intensifying competition in the text-to-video space and expanding its footprint in generative media.
Sir Michael Caine has lent his iconic voice to a University of York research project probing voice cloning and deepfake audio, helping scientists understand how synthetic speech is perceived and detected.
The legality of training AI models on copyrighted books remains a legal gray zone. Recent court rulings, licensing deals, and fair use battles are reshaping how generative AI companies source their training data.